Mechanism-embedded neural network modeling and operation optimization of a distillation unit with varying production performance

Abstract

Current challenges faced by typical industrial production processes, such as distillation unit (DU), include achieving optimum operation with varying production performance. This paper proposes a mechanism-embedded neural network (MENN) modeling and operation optimization strategy (OOS) to solve this problem. According to the physical process mechanisms, we customize the neural network structure and design key correlation constraints. This ensures that the developed MENN model can accurately reflect the process characteristics. On this basis, the operation optimization problem is formulated to adjust the top temperature of the distillation column (TTDC) to produce qualified products. The strategy, with the process physical mechanisms as the basis for the operation guidance decision, alleviates the operation blindness, especially during transition process. Industrial experiments have illustrated the proposed strategy could adapt to the different production performance. During the experiment, the proposed OOS outperformed the OOS developed by a commercial company and improved the product qualification rate.